Put your agent on the night shift
The first Friday drop. Five real jobs people handed to AI agents, posted by their authors between 9 and 14 Oct, each with the credit, the source and a prompt you can paste before you log off.
This week's five
Picked by trend score from use cases whose source is dated in the last 7 days. Every one credits its author and links the source.
Have your agent plan and book a whole trip
Grok Bot, Meta Muse, Codex · by SCOTTY BEAM · original source (12 Oct 2026)
Agents researched and booked a Lisbon trip, booked every Japan restaurant and negotiated a dinner slot in Japanese.
Build and backtest a trading bot with no coding background
Claude, Grok Bot · by Davidd Tech · original source (12 Oct 2026)
Non-developers have an agent build, backtest and run a trading strategy; claimed results are unverified.
Run Sonnet as main, Haiku explorers and Opus as advisor
Claude · by delost · original source (11 Oct 2026)
Start Claude Code with --advisor opus: Sonnet 5 codes, Haiku 5.5 subagents search, Opus is consulted only at 3 checkpoints.
Decompile a full game with a long-running agent swarm
Claude, Codex · by momo5502 · original source (9 Oct 2026)
Over 3 months and 500B tokens, worker + reviewer agents reached 99% of functions, 83% byte-exact.
Port a Windows-only app to Linux with Perplexity Computer
Perplexity Computer · by Aravind Srinivas · original source (14 Oct 2026)
Computer decompiled a keyboard app with headless Ghidra and rebuilt it in Rust for Linux, with music-synced lighting and macros.
Agent Wars: Round of 16 is open
Round one, so no results yet. Eight matchups, seeded from real use cases. Closest on paper: Cursor vs Amp, 8 use cases each. Pick your corner. The round runs 15 Oct 2026 – 21 Oct 2026; voting closes 22 Oct 2026, 00:00 UTC.
- Claude vs Manus 168:3 jobs logged
- Cursor vs Amp 8:8 jobs logged
- Meta Muse vs Pi 19:5 jobs logged
- Hermes Agent vs Antigravity 25:4 jobs logged
- Grok Bot vs Gemini CLI 53:4 jobs logged
- ChatGPT Dots vs Instinct 16:6 jobs logged
- OpenCode vs Perplexity Computer 9:7 jobs logged
- Codex vs OpenClaw 69:4 jobs logged
Power rankings, in one line
Week 42 (new use cases from 5 to 11 Oct): Claude holds #1 with 46 new, Codex holds #2, Grok Bot climbs 4 to #3, and Cursor is new on the board at #5. See the full table →
Weekly challenge: Plan a whole trip
This week: get your agent to plan a whole trip, end to end. Flights or trains, where to stay, a day-by-day plan and a budget, in one go. Booking is optional; your agent must ask you before it pays for anything.
Inspired by Have your agent plan and book a whole trip, pick 01 above.